(610h) Implementation of Genetic Algorithms to Optimize Metal-Organic Frameworks for CO2 Capture

Metal-organic frameworks (MOFs) are a class of porous crystalline materials that can be used as adsorbents to capture CO₂ from flue gas. MOFs are highly tunable; by varying their building blocks, MOFs with different chemical and physical characteristics can be designed and synthesized to reach desired properties. Computational approaches have become an important complement to experiments by directing the synthesis toward promising building blocks. In this work, we implemented genetic algorithms (GA) to search for top-performing MOFs for CO₂/N₂ separation. We aimed to provide proof of concept that using GA, we could reduce the number of molecular simulations involved by one to two orders of magnitudes, and still determined most top-performing MOFs. To implement GA, we represented MOFs as a set of discrete variables, which consists of the MOFs’ topology, nodular and connecting building blocks. All MOF structures were generated using the topological based crystal constructor (ToBaCCo 3.0) software, and their geometries were optimized using the UFF4MOF force field. Grand canonical Monte Carlo (GCMC) simulations were performed to calculate the CO₂ working capacity between the adsorption (313 K, 1 bar, CO₂:N₂ = 15:85) and the desorption conditions (313 K, 0.1 bar, CO₂:N₂=90:10), and the CO₂/N₂ selectivity at adsorption. The CO₂ working capacity and CO₂/N₂ selectivity were used as the objective functions for our implementation of GA. We used the Latin hypercube sampling to create an initial population for GA. Ranking selection methods (elitism and tournament selection) and genetic operators (crossover and mutation) were used to evolve the population. Important GA parameters, such as the number of MOFs per generation, the number of generations and the mutation probability, were determined. GA performance was tested using both a single objective function and multiple objective functions. Multiple-trajectory GA was used to avoid bad initial sampling and local convergence of GA. Cross-topology GA was used to search for top-performing structures across multiple MOF topologies. Using our implementation of GA, we could determine most of the top-performing MOFs for CO₂/N₂ separation and reduced the number of molecular simulations by more than one order of magnitude.